SLA and response quality

Scope

A weekly read of how the support operation is actually performing, entirely from Triagic's own ticket data — no integration required. Measure response and resolution timing, find the tickets that fell through, and sample reply quality. Judge the system, not individuals: the finding is "tickets arriving through channel X wait twice as long", not "this person is slow".

Procedure

1. Fix the window and the targets, and say both. Default to the last 7 days of tickets, with the previous 7 as the comparison. Then state the response targets you are measuring against. If the org has recorded targets, use them. If it has not — most have not — say so explicitly and use these as the stated working assumption: first response within 4 business hours, resolution within 2 business days for ordinary tickets and 4 hours for anything incident-shaped. Every "breach" in the report must be traceable to a target the report itself printed.
2. Build the timeline per ticket. Using the ticket data available to you, take each ticket's creation time, its current status, its last update, and its activity timeline (notes, emails, events) where one is present. Derive three numbers per ticket: time to first outbound response (the first reply that went to the requester, not an internal note), time to resolution for anything now resolved, and total age for anything still open. Where the activity timeline is missing or thin, say which tickets could not be measured and exclude them from the aggregates rather than treating "no recorded reply" as an infinite wait.
3. Distributions, not averages. Report median and p90 for first response and for resolution — a mean hides exactly the tail this checkup exists to find. Then break the same numbers down by ticket source (hubspot, zendesk, jira, slack, thread, manual) and, where the data supports it, by playbook. A channel with a much worse p90 than the others is usually a routing or notification gap, not an effort gap — check whether that source's tickets are even reaching the same queue.
4. The tail is the finding. List every ticket that breached, worst first, with its age and current status. Then look for the shapes that matter more than the individual breaches:
   • Rot — tickets in new for longer than the first-response target: nobody has picked them up at all. This is the most actionable list in the report.
   • Stalls — tickets in investigating with no activity for days: work started and stopped.
   • Silent failures — tickets with status failed: automated triage gave up and, unless someone noticed, nothing replaced it.
   • Reopens — a ticket that returned to an open status after being resolved means the first resolution did not hold, and reopen rate is a better quality signal than resolution time.
5. Sample quality, do not score everyone. Take a small sample — say ten — spread across fast and slow tickets and read the actual replies. Assess against things a reader can verify: did the first reply answer the question asked or only acknowledge it; did it ask for information the customer had already given; was the customer told what happens next and by when; did the tone match the customer's state (a frustrated second-time reporter given a template reply is a finding); did an internal detail leak into a customer-facing message. Quote at most a short excerpt, and never customer personal data. Report patterns across the sample; a single awkward sentence is not a finding.
6. Load context before concluding. Compare volume this window against the last: a p90 that doubled in a week where volume also doubled is a capacity story, and recommending "respond faster" would be the wrong answer. Say which of the two it is.
7. Close the loop with the previous run. Breaches that recur — same source, same shape, week after week — are a process finding and outrank a worse-looking one-off spike.

Finding keys

The key names the recurring problem, not the individual ticket or this week, so a channel that is still slow next week updates one ledger row and a fixed channel that slips again is caught as a regression. Key on the pattern; put the week's specific ticket ids in evidence, which is refreshed each run. A single egregious ticket may have its own key when it genuinely is the finding.

• sla:first-response:p90-breach
• source:zendesk:first-response-slow
• queue:new:rot-over-target
• status:failed:unrecovered-triage
• quality:first-reply:no-next-step
• ticket:4821:stalled-14d

Severity rubric

Severity is about customer harm and the risk of it repeating, not about the absolute number of hours.

• critical — tickets abandoned entirely: anything sitting in new or failed past several multiples of the target with no human touch, or an incident-shaped ticket that got an ordinary-ticket response time.
• high — a systematic breach rather than an outlier: p90 first response beyond target for a whole source or the whole window, or a reopen rate that says resolutions are not holding.
• medium — a meaningful minority of tickets breaching, a stalled cohort in investigating, or a repeated quality pattern in the sample (replies that acknowledge without answering, missing next steps).
• low — isolated breaches with a visible reason, small tone or clarity issues, metrics slightly worse than last week within normal variance.
• info — the week's numbers themselves, recorded as the trend baseline, and any measurable improvement worth naming.

Output guidance

Open with a one-paragraph executive summary in plain words: volume this window versus last, median and p90 first response, how many tickets breached, and the single worst pattern. Print the targets being measured against immediately after, and say explicitly whether they are the org's own or this checkup's stated assumption. Then sections for Timing, The tail (rot, stalls, failed, reopens), By source and playbook, and Reply quality — the tail section carries the actual ticket list, worst first. Close with a recommendations table (Problem | Evidence | Recommended change | Owner | Effort), process changes above exhortations to be faster. Never rank or name individual agents; every finding is about a channel, a queue, a stage, or a pattern. Say "the week met its targets" outright when it did.